Digital preservation comprises the policies, organizational arrangements, and technical activities needed to maintain access to digital information for as long as it is required. Its objects include digitized documents and photographs, born-digital publications, scientific datasets, government records, software, websites, and interactive artworks. Preservation involves more than keeping files intact: it also requires retaining the information and capabilities necessary to interpret, evaluate, and use them.
Digital information depends on changing combinations of storage media, file formats, software, hardware, and institutional services. A readable sequence of bits may become unusable if its format is undocumented, its decryption key is lost, or the software needed to interpret it no longer runs. Conversely, conversion into a current format may preserve content while changing appearance, behavior, or evidential characteristics. Digital preservation therefore combines integrity checking, documentation, technological intervention, and decisions about which properties must survive.
Libraries, archives, museums, research organizations, public authorities, and businesses undertake digital preservation for different purposes. These include maintaining cultural heritage, supporting reproducible research, meeting recordkeeping obligations, and preserving evidence of rights and transactions. The field treats long-term access as a continuing responsibility rather than a one-time transfer onto supposedly permanent media.
Scope, distinctions, and historical development
Digital preservation differs from data backup, although backups can form part of a preservation system. Backup primarily supports recovery after accidental deletion, equipment failure, or another operational disruption. It commonly restores information to a recent working state and may overwrite older copies according to a rotation schedule. Preservation instead identifies information that must persist, establishes its context and authenticity, and manages changes that could prevent future use. A successful restoration does not establish that a file is understandable, complete, or the version intended for permanent retention.
Preservation also differs from digitization. Scanning a manuscript creates a digital representation; it does not itself ensure that the resulting images will remain accessible. A digitization project may need to preserve image masters, structural relationships between pages, descriptive records, capture settings, and information about the physical original. Born-digital materials pose a different problem: an email collection or interactive artwork may have no stable physical counterpart from which it can be recreated. A printed copy can preserve some visible content while losing attachments, executable behavior, or machine-readable structure.
The field developed from earlier work in computing, archives, and scientific data management. A prominent early case was the United States’ 1960 population census, whose computer-readable records required continuing management as storage technologies changed. By the 1990s, networked publishing and growing institutional dependence on electronic records made technological obsolescence a widespread archival concern. The 1996 report Preserving Digital Information, produced by a task force associated with the Commission on Preservation and Access and the Research Libraries Group, emphasized organizational responsibility and a distributed infrastructure rather than a purely technical solution.
International frameworks consolidated during the following decade. The Consultative Committee for Space Data Systems published its Open Archival Information System reference model in 2002; it became ISO 14721 in 2003. Also in 2003, UNESCO adopted the Charter on the Preservation of Digital Heritage, recognizing that digital resources could be lost through rapid technological change and inadequate institutional preparation. These developments helped establish preservation as a lifecycle activity beginning before acquisition.
“Long term” is not a fixed number of years. It means a period sufficiently extended for technological changes, changes in users, or loss of contextual knowledge to become relevant. A business record retained for ten years may require active preservation, while a frequently used file in an unchanged environment may need little intervention over the same interval. Requirements depend on the information, its dependencies, and the people expected to use it.
Selection, appraisal, and transfer into custody
Preservation begins with decisions about what to retain. Digital storage capacity does not eliminate appraisal: collections can contain redundant copies, obsolete working files, personal information, undocumented exports, and material outside an institution’s mandate. Selection may consider legal obligations, research value, uniqueness, representativeness, acquisition costs, and the feasibility of providing meaningful future access. Decisions also operate at different levels. An archive may preserve an entire organizational email account, selected correspondence, or a defined set of records extracted from a larger system.
The unit of preservation is not necessarily a single file. A research dataset may depend on a codebook, software scripts, calibration records, and descriptions of missing values. A digital publication may consist of text, images, fonts, stylesheets, and supplementary data. A database may require its schema, constraints, stored procedures, documentation, and a record of the extraction process. Identifying these relationships prevents the preservation of apparently complete files that are unusable without missing components.
Transfer into a repository is usually governed by an agreement specifying the material, responsibilities, rights, and expected documentation. An inventory or manifest can record filenames, file sizes, identifiers, and checksums. The receiving institution compares the delivered material with this inventory and records discrepancies. Transfers may arrive through network services, external drives, optical discs, or complete disk images. For older or fragile media, capture may require specialized readers and procedures that minimize changes to the source.
Ingest commonly includes quarantine, malware scanning, file identification, validation, extraction of technical characteristics, and creation of preservation records. These operations answer different questions. Format identification estimates what kind of file is present, often by examining internal signatures rather than filename extensions. Validation tests conformance to a format specification. A valid file can nevertheless be incomplete or misleading, while an invalid file may still render successfully in common software. Malware scanning likewise reduces a particular security risk without establishing archival authenticity.
Appraisal and processing decisions require documentation because they shape the surviving record. Removing duplicate files may reduce costs, but identical files in different directories can carry different contextual meanings. Renaming files may improve consistency while breaking links or obscuring original organization. Removing sensitive information creates a derivative that must be distinguished from the received object. Some repositories retain original submissions alongside normalized or redacted versions, documenting the relationship between them.
Early collaboration with creators can substantially improve preservation readiness. Export facilities, stable identifiers, clear licensing, and documented formats are easier to establish while a system is operating than after it has been decommissioned. This approach connects preservation with data management and recordkeeping, while retaining the distinction between managing information for current operations and accepting responsibility for its future intelligibility.
Reference models, metadata, and authenticity
The Open Archival Information System reference model, commonly called OAIS, provides a conceptual vocabulary for preservation. Its central participants are producers, management, and consumers. A repository must identify a “designated community”: the intended group of users whose knowledge and capabilities determine what supporting information is necessary. A dataset intelligible to specialists familiar with an instrument may be unintelligible to later researchers without documentation of units, processing methods, and experimental conditions.
OAIS distinguishes submission, archival, and dissemination information packages, usually abbreviated SIP, AIP, and DIP. A producer supplies a submission package; the repository establishes an archival package for managed retention; and users receive dissemination packages appropriate to access needs. These are conceptual roles, not mandatory file layouts or software products. An archival package may be implemented through files, storage objects, database records, and relationships distributed across several systems.
A crucial OAIS concept is representation information: information needed to interpret the stored data. For a simple text file, this may include its character encoding. For a scientific array, it may include dimensions, units, coordinate conventions, and the meaning of special values. Representation information can itself require explanation, creating a network of dependencies. The designated community limits how far that network must extend: a repository need not teach every user all relevant mathematics, but it must state what knowledge it assumes.
Metadata supports identification, management, interpretation, and access. Descriptive metadata helps users discover an object; structural metadata records relationships such as page order; technical metadata describes properties such as image dimensions or compression; and administrative metadata includes rights and management information. PREMIS, whose first data dictionary was released in 2005, focuses on preservation metadata organized around objects, events, agents, and rights. A format migration can therefore be recorded as an event involving an input object, an output object, a software agent, and an outcome.
Authenticity concerns whether an object is what it purports to be. It is supported by documented provenance, reliable custody, controlled procedures, and evidence of changes. Integrity is narrower: it concerns completeness and freedom from unintended alteration. A checksum can show that a file matches an earlier state, but it cannot prove that the earlier state was genuine or that a document’s assertions are true. Digital signatures can establish additional evidence about signing and alteration, but their long-term interpretation may require certificates, validation records, timestamps, and knowledge of the signing environment.
Preservation documentation must also survive independently of individual staff members or a particular repository application. Persistent identifiers, exportable records, explicit relationships, and documented vocabularies support continuity. A technically intact collection can become effectively lost when its identifiers no longer resolve or the only explanation of its organization disappears with an obsolete catalog system.
Storage, replication, and integrity control
The physical foundation of preservation is reliable retention of bits, often called bit-level preservation. Storage media fail through component faults, wear, environmental damage, handling errors, and failures in controllers or software. Silent corruption is especially significant because information may change without producing an obvious read error. No single storage technology eliminates these risks. Magnetic disks, tape, optical media, and solid-state devices have different operational characteristics, but all require management and eventual replacement.
Repositories use cryptographic hash functions to generate compact values for checking file integrity. If an object is represented by , its recorded digest is . A later check computes from the currently stored object and compares it with . A mismatch indicates that the bytes differ or that an error has affected the checking process or recorded reference. A match provides strong evidence of bit-level continuity when an appropriate hash function and trustworthy baseline are used; it does not establish semantic correctness.
Fixity checking becomes useful only when paired with response procedures. The repository must decide how frequently to check objects, how to investigate failures, and which copy can serve as a repair source. Checks performed during ingest, transfer, storage migration, and retrieval can detect errors introduced at different stages. Reference digests and logs need protection as well: an attacker or faulty process that changes both an object and its recorded checksum can defeat an inadequately separated control system.
Replication reduces dependence on any single copy. Preservation copies are often separated geographically and administratively to limit correlated failures, such as flooding, a shared software defect, or a compromised administrator account. Merely placing several copies in the same building or under the same credentials provides less independence. Multiple copies also require comparison and repair mechanisms; otherwise, a corrupted version may spread through synchronization or remain unnoticed across all replicas.
Storage architectures may combine disk, tape, and cloud computing. Online storage favors rapid access, while nearline and offline systems can reduce some costs or exposure to network attacks. Cloud object storage offers managed redundancy and durability features, but a provider’s durability claim does not cover every preservation risk. Account deletion, configuration errors, loss of encryption keys, provider dependency, retrieval charges, and legal jurisdiction remain relevant. Export capability and tested transfer procedures matter when a repository must change service providers.
Storage planning also accounts for overhead beyond the original files. Replicas, metadata, logs, indexes, temporary processing space, and derived access copies can substantially increase capacity requirements. Media refreshment transfers content to newer storage without necessarily changing file formats. Its success is established through verified copying and reconciliation of inventories, not simply through a report that the new device is functioning.
File formats, migration, and significant properties
A file format specifies how information is organized and encoded. Its preservation prospects depend on more than age or popularity. Relevant factors include the availability of specifications, implementation diversity, external dependencies, support for embedded metadata, and the ability to validate conformance. An openly documented format is generally easier to analyze and reimplement than an undocumented proprietary one, but openness alone does not guarantee that every feature will remain usable.
Repositories distinguish format identification from risk assessment. A rarely encountered format may be well documented and easy to decode, while a widely used format may depend on remote services or proprietary components. Registries such as the United Kingdom National Archives’ PRONOM provide identifiers and technical information that assist identification and planning. Tools including DROID use format signatures to recognize files. Such tools support decisions but do not replace examination of how a particular collection uses a format.
Format migration converts information from one encoding or format into another that is expected to remain usable. Examples include converting a proprietary word-processing document into a documented preservation format, exporting database tables into documented structures, or transcoding audiovisual files. Migration can also occur between versions of the same format. Because conversion software interprets and rewrites information, the output is normally a new object whose relationship to the source must be recorded.
The concept of significant properties helps specify which characteristics must survive. For a textual record, the words, reading order, and pagination may matter; for a spreadsheet, formulas and recalculation behavior may be essential; for an artwork, timing and interaction may outweigh pixel-for-pixel similarity. These requirements are contextual rather than universal. Converting a spreadsheet into a static image preserves appearance at a particular moment but removes formulas, editable cells, and computational behavior.
Validation of migration requires tests related to these properties. A repository may compare page counts, extracted text, image dimensions, color information, embedded attachments, or calculated results. Automated checks can reveal structural discrepancies, while visual or functional review may be necessary for representative or high-risk objects. Conversion logs alone are insufficient: software can report success while substituting fonts, omitting unsupported features, or changing date interpretations. Retaining the source allows later reassessment.
Preservation-oriented formats reduce particular risks without solving the entire problem. PDF/A, first standardized in 2005, constrains PDF features to support long-term document preservation, including restrictions on external dependencies and encryption. It does not establish the accuracy of a scan or guarantee that every document is accessible to assistive technology. Similarly, lossless compression preserves the encoded input exactly when correctly decoded, whereas repeated lossy transcoding can introduce cumulative changes. Format policies therefore distinguish preservation masters, access derivatives, and the purpose each serves.
Emulation, software, and complex digital objects
Migration is not always the best way to preserve software-dependent material. An interactive program may rely on a particular processor, operating system, display behavior, or peripheral. Converting its files cannot necessarily reproduce its execution. Emulation addresses this problem by recreating aspects of an older computing environment on a newer system. Instead of transforming the preserved object into a current format, it aims to provide an environment in which the object can operate.
An emulator may reproduce a processor’s instruction set, the behavior of an entire machine, or selected interfaces. The preserved environment can include disk images, an operating system, applications, fonts, configuration files, and user data. Virtualization is related but often relies on compatibility between host and guest architectures and on facilities of the underlying hardware. Neither approach is automatically permanent: emulator code, configuration knowledge, and host environments themselves require maintenance.
The 1986 BBC Domesday Project illustrates the interdependence of content and technology. Its interactive resources combined computer software with material stored on specialized video discs. Later recovery and access efforts needed to address both the digital material and the original system’s interaction with its media. Similar issues occur in video games, educational software, digital art, and engineering applications. Preserving files alone may leave essential behavior inaccessible.
Capturing an executable environment does not guarantee faithful experience. Timing differences can affect games and simulations; a modern display may change color and motion; input devices may lack equivalent controls. Networked software may depend on authentication servers, licensed services, or live communities that cannot be reconstructed from the client alone. In such cases, documentation, screen recordings, interviews, and records of interactions can preserve evidence of experience without reproducing it completely.
Scientific and business software introduce additional requirements. Retaining source code supports inspection and possible reconstruction, but rebuilding may require compilers, libraries, build scripts, and precise versions of dependencies. Binary executables may preserve an operational state that is difficult to reproduce from source. Containers can package applications and dependencies, yet still rely on a host kernel, processor architecture, and external services. Reproducible builds and documented environments reduce uncertainty but do not remove every dependency.
Execution also creates security and rights problems. Legacy software may contain vulnerabilities, and archived programs may be malicious or behave unpredictably. Repositories can use isolation, restricted networking, and controlled access to reduce these risks. Software licensing may constrain copying or public execution even when preservation is technically feasible. The resulting archival object may therefore comprise several coordinated elements: original files, executable environments, technical documentation, access restrictions, and recorded evidence of behavior that cannot be recreated.
Access, web archiving, and changing information
Preservation and access are connected but not identical. A repository may retain material that cannot immediately be made public because of copyright, confidentiality, donor agreements, or restrictions on personal information. It may also provide access through derivatives rather than preservation masters. An audiovisual master can remain unchanged while users receive streamed versions; a fragile software environment may be offered through a controlled remote session rather than as a downloadable system image.
Discovery requires stable identification and adequate context. Catalog records, collection descriptions, persistent identifiers, and search indexes help users locate material, but these services should not become the only record of what exists. Access systems are frequently replaced, so repositories need to preserve the relationships between descriptive records and stored objects. A citation should identify a particular version or state where the distinction affects interpretation.
Web archiving demonstrates the difficulty of preserving information assembled dynamically. A page on the World Wide Web may combine an initial document with stylesheets, images, scripts, video, and responses from remote services. A crawler can follow links and collect server responses, but it may miss content requiring login, forms, user interaction, or location-specific requests. Browser-based capture can execute scripts and reveal additional resources, although it remains a record of selected interactions rather than an exhaustive copy of every possible state.
The WARC format, standardized as ISO 28500 in 2009, provides a structure for storing web capture records, including request and response information and related metadata. Replay systems use captured resources to reconstruct archived pages. A capture timestamp does not imply that every component originated at exactly that moment: a crawl collects resources over an interval, and some components may be absent or drawn from different captures. Replayed appearance therefore needs to be distinguished from an exact historical reconstruction.
Other changing systems require different capture models. A database snapshot records a state at a particular point, while transaction records can document changes over time. An exported social media account may preserve posts and attachments but omit platform-wide relationships, recommendation behavior, or deleted interactions. For email, headers, attachments, message identifiers, and threading relationships can be as important as visible message text. Preservation boundaries must identify whether the object is an individual record, a collection, an application, or evidence of a broader service.
Access also includes usability and accessibility. Searchable text, captions, transcripts, and structured navigation can make preserved material more useful, but generated derivatives may contain errors or impose interpretations absent from the original. Repositories therefore distinguish original content from later enhancements and retain documentation of how those enhancements were produced. Continued access involves maintaining both the preserved information and the systems through which users can understand its provenance and limitations.
Governance, trustworthy repositories, and sustainability
Digital preservation depends on organizational continuity as much as technical design. A preservation policy defines responsibilities, collection scope, retention objectives, permitted actions, and the relationship between preservation and access. Operational procedures translate these commitments into ingest checks, storage management, integrity monitoring, incident handling, and documentation. Clear ownership matters because preservation tasks often cross archival, legal, financial, and information-technology departments.
A trustworthy digital repository must provide evidence that it can meet its stated commitments. ISO 16363, published in 2012, establishes an audit and certification framework addressing organizational infrastructure, digital object management, and infrastructure and security risk management. The CoreTrustSeal framework provides another form of repository assessment. Such frameworks evaluate documented capabilities and practices; certification does not guarantee that no object will ever be lost or that every preservation strategy is appropriate for every collection.
Risk management considers both likelihood and consequence. A rare event affecting an irreplaceable collection may warrant more attention than a frequent, easily reversible fault. Repositories track risks such as unsupported formats, incomplete inventories, staff dependency, failed suppliers, and legal uncertainty. Measures overlap with information security and disaster recovery, but preservation adds concerns about meaning, authenticity, and technological interpretability. Strong security that permanently blocks legitimate future access can itself defeat preservation objectives.
Legal and ethical obligations shape retention and use. Intellectual property rights may affect copying, migration, emulation, and distribution. Data privacy requirements may constrain the retention or disclosure of personal information. Applicable exceptions, obligations, and archival provisions vary by jurisdiction. Institutions also encounter culturally sensitive materials and restrictions established through agreements with communities or depositors. Access controls and documented decisions allow preservation to continue without treating unrestricted publication as its necessary outcome.
Costs extend beyond storage purchases. Appraisal, rights work, documentation, software maintenance, security, integrity checking, and access services require continuing labor. Costs can rise sharply when material depends on specialist hardware or poorly documented software. Conversely, routine workflows, shared infrastructure, and well-documented formats can reduce repeated effort. Preservation planning therefore considers the entire lifecycle, including eventual migration to another institution or service, rather than treating initial deposit as the main expense.
Environmental sustainability introduces further trade-offs. Replication, frequent checking, and rapid access consume energy and require equipment, while unnecessary deletion can make later recovery impossible. Decisions about storage tiers, checking schedules, redundant derivatives, and the number and independence of copies affect both resource use and resilience. There is no single copy count or refresh interval appropriate to every collection. Sustainable stewardship depends on explicit preservation objectives, measured risks, and documented choices about which costs and dependencies an institution can maintain over time.